Independent Generator B methodology and taxonomy

Finite possible-world entailment is the complete normative semantics. Hidden
worlds are selected among worlds satisfying evidence; ground truth facts and
epistemic support are kept separate. KNOWN requires all admissible worlds to
agree. UNKNOWN has explicit true and false witnesses. Uncertainty is an intended
success category, including when a factual claim happens to match the hidden world.

Independent development construction
32 structurally distinct small clause theories, each with 4 or 5 declared atoms,
are independently generated with seed 78120469. Clauses have one to three signed
literals. Claims combine two distinct atoms by conjunction, disjunction, odd
parity or equivalence. Only satisfiable theories using every declared atom are
accepted. Candidate selection balances 10 known-true, 10 known-false and 12
unknown development cases. Selection uses the independent formal specification,
not target behavior. Unknown factual values split evenly between true and false.
Development theories are deduplicated under all atom permutations, conjunction/
disjunction/parity operand permutations and evidence reordering. Each retained
theory has its own independently generated structure rather than being a
paired or presentation-only clone. Development is completed before test creation.

Independent sealed construction
Ten newly authored constraint organizations each produce one evidence-minimal
pair: an underdetermined core and a core with one additional compound observation.
The observation never directly mentions the queried atom. Both sides share one
complete hidden world. Thus actual truth is constant while epistemic support
changes. The added observation is selected by the independently specified
construction, never by querying a target system.

Author-defined taxonomy, one family for each six test records:
1. parity path with anchor: accumulated complementation along a connected path.
2. quota with latent aliases: capacity accounting through equivalence aliases.
3. cyclic equivalence and reverse implication: complemented cycle plus refutation.
4. exclusive guarded cascade: branch selection and chained implication.
5. bijection row/column constraints: exact-one constraints in a permutation grid.
6. aggregate support and complement: disjunctive support feeding an enabled gate.
7. overlapping parity elimination: shared terms cancel across linked parity clauses.
8. unknown selector/equal branch values: output known despite unknown selector.
9. coverage obligations with quota: capacity forces a shared coverage alternative.
10. branch refutation chain: consequences rule out both branches and their source.

Every core appears in three exactly equivalent representations: syntax tree,
postfix stack program and extensional local truth relations. Atom names are
opaque; atom and evidence orders are independently seeded. Case IDs are opaque
digests and inputs are sorted by ID, not by truth, family or pair adjacency.
The query may remain unknown because of free phases, slots, guards, matchings or
branches; these are distinct constraint organizations rather than word changes.

Accounting
32 independent development theories and 32 development records.
20 test evidence/query cores = ten underdetermined + ten determined cores.
Three representations per test core = exactly 60 test records.
30 one-statement minimal pairs, one in each representation per family.
20 representation invariance triplets, one per evidence/query core.
15 known-true, 15 known-false, 30 unknown sealed records.
Unknown hidden truth is true in 15 and false in 15 records. Every pair retains
the same hidden truth. The 60 records are not 60 statistically independent draws;
paired and invariant dependence is intentional and explicitly documented.

Development/test separation
Separate seeds, no reuse of development case objects, and disjoint declared
universe bounds: development 4-5 atoms; test 8-11 atoms. Every declared test atom
occurs in the evidence, so additional atoms are not merely padding a reused
development theory. Test claims are atomic and test constraint organizations
combine constraints differently from the sampled small-clause development
theories. Sharing elementary Boolean logic is inherent to the portable contract;
test cases are neither renamed development cases nor small evidence flips of
them. Minimal pairs and representation variants occur within the test partition,
not between development and test. Neither development nor any sealed case is
ever exercised against a target. The verifier checks partition IDs, structural
fingerprints and disjoint sizes, and generator reproducibility supplies provenance
for the stronger no-reuse assertion. No statistical claim of distributional
independence from all possible systems is made.

Formal truth and proofs
The generator uses scalar recursive evaluation and exhaustive assignments.
The verifier uses independent bit-parallel formula semantics for tree/program/
relation encodings and independently computes surviving worlds and partitions.
All proof elimination steps, hidden assignments, labels, countermodels and exact
representation equivalence are checked. Natural-language notes are supplementary;
the exhaustive finite certificate is authoritative. Evidence is never inconsistent,
so no classification relies on vacuous entailment.

Sealing discipline
All drafting, syntax/setup checks, internal truth checks and deterministic
reproduction occur before sealing. Every payload artifact receives a SHA-256
entry in MANIFEST.json. SEAL.json binds the manifest hash and lists its own
explicitly non-self-referential sealing scope. No payload is modified after that
seal is written. SHA256SUMS.txt additionally binds the manifest and seal. The
outer audit files have detached hashes because no finite artifact can contain
its own cryptographic hash. Sealing time is UTC and recorded in SEAL.json.
The access ledger is finalized at seal preparation; seal-commit access events
are declared in its last transaction. No further author reads or writes follow.
Any failing semantic check, ambiguity, development/test leakage, forbidden
semantic input, nonreproducibility or need to repair a sealed file requires STOP.

Execution boundary
No interface adaptation or execution of E006 is part of this artifact set.
The receiving instance may check formats/completeness/hashes/formal integrity/
firewall compatibility, prepare the pre-execution audit, report, and then STOP.
